The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructions direct users to download binary releases and clone source code from an untrusted third-party GitHub repository (Ronish17/redbull-archive-cli) instead of the vendor's own infrastructure.
[REMOTE_CODE_EXECUTION]: The installation process involves downloading a pre-compiled binary (nebulavault-linux-x64.tar.gz) and executing it locally, which allows for the execution of unverified code.
[COMMAND_EXECUTION]: The post-processing feature in the configuration settings allows the definition of shell commands (e.g., using ffmpeg) that interpolate variables like {input}. This presents a command injection surface if media filenames or associated metadata are manipulated by an attacker.
[DYNAMIC_EXECUTION]: The skill includes a plugin system that allows loading and executing arbitrary JavaScript files from the local filesystem, which can be leveraged to run malicious logic within the Node.js or Deno environment.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external sources, including RSS feeds and streaming URLs.
Ingestion points: Data enters the system via the feed add command and media metadata extraction during downloads.
Boundary markers: There are no explicit instructions or delimiters to prevent the agent from following instructions embedded in the processed metadata or RSS content.
Capability inventory: The skill possesses extensive capabilities including file system writes, network access, and shell command execution via post-processing.
Sanitization: No evidence of sanitization or validation is provided for data retrieved from external feeds or media streams before it is used in further operations.

nebulavault-media-downloader